Unified Optimal Analysis of the (Stochastic) Gradient Method
Machine Learning
2019-12-24 v2 Numerical Analysis
Numerical Analysis
Optimization and Control
Machine Learning
Abstract
In this note we give a simple proof for the convergence of stochastic gradient (SGD) methods on -convex functions under a (milder than standard) -smoothness assumption. We show that for carefully chosen stepsizes SGD converges after iterations as where measures the variance in the stochastic noise. For deterministic gradient descent (GD) and SGD in the interpolation setting we have and we recover the exponential convergence rate. The bound matches with the best known iteration complexity of GD and SGD, up to constants.
Cite
@article{arxiv.1907.04232,
title = {Unified Optimal Analysis of the (Stochastic) Gradient Method},
author = {Sebastian U. Stich},
journal= {arXiv preprint arXiv:1907.04232},
year = {2019}
}
Comments
11 pages, version 2 fixes typos and case distinction in the proof